agentleFS
Sign inStart free
yannick-bottinoyannick-bottino/consulting-claude-skills/skills/folder-analyzer-optimizer/CLAUDE.md

consulting-claude-skills / folder-analyzer-optimizer

CLAUDE.md0 starsChanged 53 days agoMITView on GitHub

Optimisation du skill folder-analyzer-optimizer pour reduire la consommation de tokens de Claude lors du parsing de dossiers documentaires, a qualite constante. Construire un harness d'auto-optimisation inspire du pattern autoresearch (Karpathy) : - Boucle autonome qui itere sur les parametres du pipeline de parsing - Mesure tokens consommes + qualite du parsing sur un dataset de reference - Keep/discard chaque experience, log dans results.tsv - Max 20 experiences Base documentaire/ — 125 fichiers : - 109 PDFs (majoritairement attestations thermiques RT2012/RE2020)…

The file

We split this CLAUDE.md into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep/discard chaque experience, log dans results.tsv

    line 12
  2. Distinctive:

    Keep/discard, log dans results.tsv

    line 36
  3. Distinctive:

    Run complet sur 125 fichiers avec la meilleure config

    line 39

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

Folder Analyzer Optimizer

Projet

Optimisation du skill folder-analyzer-optimizer pour reduire la consommation de tokens de Claude lors du parsing de dossiers documentaires, a qualite constante.

Objectif actuel

Construire un harness d'auto-optimisation inspire du pattern autoresearch (Karpathy) :

  • Boucle autonome qui itere sur les parametres du pipeline de parsing
  • Mesure tokens consommes + qualite du parsing sur un dataset de reference
  • Keep/discard chaque experience, log dans results.tsv
  • Max 20 experiences

Dataset de reference

Base documentaire/ — 125 fichiers :

  • 109 PDFs (majoritairement attestations thermiques RT2012/RE2020)
  • 12 XLSX
  • 2 PPTX
  • 1 DOCX
  • 1 .msg

Approche

Hybride : template detection + boucle d'optimisation (approche C du design)

Phase 1 : Baseline & Clustering

  • Parser les 125 fichiers avec le skill actuel, instrumenter tokens + qualite
  • Clusterer les PDFs par template (fingerprint structurel via PyMuPDF, 0 tokens)

Phase 2 : Boucle autoresearch (max 20 iterations)

  • Agent modifie config.py (seuils, DPI, routing, template detection, format output)
  • Execute optimize.py sur un subset representatif (~15 fichiers)
  • Mesure delta tokens vs baseline, verifie qualite (proxy metrics + spot-check LLM)
  • Keep/discard, log dans results.tsv

Phase 3 : Validation & Cleanup

  • Run complet sur 125 fichiers avec la meilleure config
  • Spot-check LLM sur 10%
  • Appliquer gains au SKILL.md
  • Supprimer tout le harness d'optimisation

Spec detaillee

docs/2026-03-31-autoresearch-optimization-design.md

Contraintes

  • Le harness est temporaire — supprime apres optimisation
  • Qualite = mix metriques proxy (0 tokens) + spot-check LLM (echantillon)
  • Le SKILL.md existant n'est pas modifie pendant l'optimisation, seulement a la fin
  • program.md = edite par l'humain. config.py = edite par l'agent. Rien d'autre

Stack technique

  • Python 3 : pymupdf, python-docx, python-pptx
  • Metriques proxy : word count, table count, heading count (0 tokens)
  • Clustering : fingerprint structurel PyMuPDF + similarite (0 tokens)

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free